OSHW TO247 IGBT watercooled laminated half-bridge

HighHopes said:
did u use seeeds to fab the board too? 1oz is good pricing, but 2oz is more than 10x?
You mean the controller board? That was made in http://www.ourpcb.com, they are fast, a bit more expensive though but they can do whatever I come up with, I've been working with them for ages.

I just checked again at seeeds, this half bridge board is now $96 @1Oz and $126 @2Oz for qty=5. Their prices seem to change fast, go figure.

Something I didn't realize is that with the extra copper thickness they derate their minimum track clearance/thickess. The board was laid out at 8mil clearance / 10mil min track width, but at 2Oz seeed only support 10/10. Thats a bummer, I'm not willing to change the layout for that. The last board I sent to ourpcb was 6/8 mils at 2 Oz.

Itead studio also had fairly cheap pcb prototyping, but for this board I might send the job to ourpcb. I don't like the qty=5 to qty=10 jump, I would prefer to just get 6 pcbs, and with ourpcb the boards could jump into our regular pcb shipments, which to me that counts as free fedex shipment.
They also separate the tooling cost from the pcb cost, so if I order now it could be $90 tooling plus $10 per board. If the next guy wants a pcb he only needs to pay the $10 per board.

I'm busy at work now for some days, before sending this gerbers I want to have the copper sheets designed in case something new comes up. I'm also counting on a small 2 layer board to hold a LEM current sensor and a differential opamp so the current signal is fully differential from the sensor. If the distance between the current sensor and this board is not too long it could be a flexible pcb.
 
wait till u see the price of the flexible PCB ;P
 
Its around 90usd to make an FPC prototyping batch, I made a few hundred, it ends in 1usd each. Adds a bit to the cool factor and you get differential traces very close to each other on top of a copper plane 0.15mm away. And you can solder smd components and use pth vias. Did I say I hate wiring?
Its only an option, its for sure doable with a 10usd pcb plus wires.
 
Does anyone know a good and fast delta sigma current sensor for ~1000amps?

I've been toying with the idea since day 0.
Pros:
* if the voltage and current signals were delta sigma encoded I could replace 13 opamps by a single fpga in the controller
* Fully digital signal path. Zero noise picked up from the sensor to the microcontroller
* Error detection. Most delta sigmas I've seen implement a way to check if the sensor is not ok. (a stream of 0's means error for example)
* For phase voltage measurement its already here, just change a part# and its delta sigma instead of analog differential. Same footprint and circuit.
* Having an fpga I would remove the discrete logic pwm overlap circuit.

Its a total of 10 ICs plus countless rc filters replaced by a single QFN48 IC.

Cons:
* Its been 5 years since the last time I developed fpga stuff, could take me a while to get this simple hdl done
* Sinc3 filter response time. Right now I have a 4usec response time and I'd like to keep it that way without losing resolution.
* Minor code change to read a DMA-filled register instead of ADC-filled register. fpga would write the microcontroller ram using dma, no cpu involved.
* Apparently no off the shelf current sensors available

So coming back to the original question, are there suitable delta sigma current sensors?
So far the best approach I could find is to build my own using something like this 13usd resistor http://www.isabellenhuette.de/fileadmin/content/praezisions-leistungswiderstaende/BAS.PDF
plus the same isolation amplifier used in the phase voltage...
 
Hi marcos,

have a look at the AMC1304L05. +-50mV input Range with 80dB SNR. But the shunts needed should be big ones :). Its a pitty LEM has no sensors with sigma delta interface.

On the FPGA need: I use these AMC1304 with an STM32F767 which has the sinc filters on chip, no need for an FPGA. But my currents are in the range of 100 - 300A.

regards

Tec
 
What about a simple hall effect sensor? I use a csla2en sensor it's + - 950 amp AC/DC 3 us response time
 
have a look at the AMC1304L05. +-50mV input Range
Almost exactly, I was aiming at the AMC1303, thats the one footprint compatible with this layout.
But the shunts needed should be big ones
Big and not easy to order from a common supplier like digikey afaik. LEM uses hall sensors witch are already isolated and already have some hysteresis and lag, so adding a delta modulation on top of that seems like a bad choice.

Looks like shunt resistor is the only way to go for a digital current measurement. At 50mv 1000A its 50W per phase shunt. Hmmm.

On the FPGA need: I use these AMC1304 with an STM32F767 which has the sinc filters on chip, no need for an FPGA. But my currents are in the range of 100 - 300A.
I hear you there. I almost use that family for a job that needed the delta sigma filter built in. In this case I can't steer the vesc project (the controller I plan to use) to migrate from STM32F4 to STM32F7, but it was a good call, thanks.
 
Arlo1 said:
What about a simple hall effect sensor? I use a csla2en sensor it's + - 950 amp AC/DC 3 us response time
Thats similar to what I'm using now, its been working for a couple of months. So far this is only an idea that would lower noise, tolerances, and parts count/cost plus adding fault detection in exchange of more development hours with the fpga code...
 
marcos said:
Arlo1 said:
What about a simple hall effect sensor? I use a csla2en sensor it's + - 950 amp AC/DC 3 us response time
Thats similar to what I'm using now, its been working for a couple of months. So far this is only an idea that would lower noise, tolerances, and parts count/cost plus adding fault detection in exchange of more development hours with the fpga code...
Why don't we start a current sensor thread. I am about to. I need 2000amp current and soon. But I also use a melixis current sensor on the DC input wire to the controller for my gauges. It's hard to get a sheild that has 0 magnetic memory....
 
Why don't we start a current sensor thread. I am about to.
Go for it and I'll follow.
 
i've never used shunt for high power. i don't believe you'll get the SNR necessary, especially at low to mid range amps (at very low, probably have to turn current loop off). anyway, i did look into it a few months ago for a high amp project. i'll post what i found in your other thread which relates to shunt resistors.

combining an FPGA with simple uC makes a lot of sense for motor control project but its not really necessary to get the basic thing working well. i have done this before including a CPLD instead of the discretes for overlap avoidance. but i did this because FPGA counts as hardware, not software, and there are certain advantages to that for certification reason. for one-off build of personal use, dunno if its worth your time to develop. $0.02
 
Ok I started a current sensor thread here. https://endless-sphere.com/forums/viewtopic.php?f=30&t=89911
 
HighHopes said:
for one-off build of personal use, dunno if its worth your time to develop.
Maybe I'm trying to get away from one-off build territory :wink:

Having an fpga as supervisor (and digital filter for these digital signals) would make safer to use a very non certified brain. Thats the thing with my controller, its a industrial non automotive mcu running non certified garage tested firmare. Honestly I don't know if a fpga would ease the certification path in this case. If it doesn't I'd lose a big reason to get involved in the fpga design.
 
Reviewing the board this week I noted a couple of things:

1) If a booster circuit with external pnp+npn is used to increase gate drive current, the ISO5852 built-in soft turn-off doesn't work, so I had to add an R+C. (tiduc70a.pdf) This way the turn-off during a desat fault would extend for around 3us. Basically the IC expects some capacitance in the IGBT gate to build the delay but the small pnp bjt has almost no capacitance, so we add those 10nF that resembles an IGBT gate.

Good thing I saw this, it would have detected desat but kill the igbt due to turn off overvoltage.

2) Added igbt active clamping for safe turn-off during fault currents.
Which is not the same thing as a desat fault turn-off. In case there is a very high current and the gate driver is commanded to turn off just *before* a DESAT event, the IGBT would turn off fast as usual, the Vce overvoltage would be high and no soft turn-off would happen (its only triggered by a desaturation event). This new hardwired clamping will make the IGBT avoid overvoltage conditions, usually caused by interrupting a large current flow.

gate driver mods.png

Also changed the boost transistors for an AEC-Q101 qualified part.
 
Marcos, I have been to busy to read this whole thread. But I started yesterday.

Great work so far. I am also looking at something to do slow turn off and would like something watching voltage on the collector.
I was looking at the infinion driver which claims to do this... But doing it with discrete parts would also be cool.
The Avego ACPL-333j I use at the moment just says slow turn off during desat but I think my driver would be a little more reliable with something that watched the voltage of the collector during desat. I am always pushing the limits and with things cranked up to 11 its not to hard to cause a failure.. When things are turned down to a more realistic limit its not so bad.

Will finish reading the thread asap.

-Arlin
 
Welcome along!

What do you mean by watching collector voltage during desat?
Desat detection means you are already measuring the collector voltage+gate voltage+blanking cap.

In the last post I mentioned active clamping using a zener to clamp the collector voltage. I took that from HighHope's gate driver and some app notes.
Google for igbt active clamping, it will keep ypur Vce voltages under control, but beware, zeners are far from ideal, look at their curves and you will see.

One thing I like TI stuff is their capacitive isolation, its very fast, low jitter, high dVdt. SiC worthy for sure. High end infineon drivers look great but I sense its going to be difficult to order and will require some hardened firmware to deal with their SPI management.

Keen to hear your thoughts.
 
So to start with I think I have suffered from to much voltage rise on VCE during desat events. So If I had a better setup to watch that VCE spike and control it rather then the ACPL333j just shutting it off what it thinks is slow... I think I would move from 99.9 % reliable to 99.9999% ;)

I first discovered it on page 29-31 on this doc https://www.infineon.com/dgdl/Infineon-AN2014_08_Hybrid_Kit_for_HybridPACK2-AN-V02_04-en.pdf?fileId=db3a30432239cccd01225a7cfb166c8a
 
I first discovered the zener clamping here
http://ivanbennett.com/forum/index.php?PHPSESSID=1l2ts20u38pal7a2j91atk8mc3&action=dlattach;topic=112.0;attach=269

Later I found its called active clamping and Power Integrations did some improvements over the basic concept, but I liked that basic concept so I added it to the schematic.

Wanna give it a try? 2 cheap parts, rather big in size though (SMC package)
https://octopart.com/search?q=SMCJ440CA (I think its ok for your bus voltage)
https://octopart.com/search?q=ES2J-LTP
 
marcos said:
I first discovered the zener clamping here
http://ivanbennett.com/forum/index.php?PHPSESSID=1l2ts20u38pal7a2j91atk8mc3&action=dlattach;topic=112.0;attach=269

Later I found its called active clamping and Power Integrations did some improvements over the basic concept, but I liked that basic concept so I added it to the schematic.

Wanna give it a try? 2 cheap parts, rather big in size though (SMC package)
https://octopart.com/search?q=SMCJ440CA (I think its ok for your bus voltage)
https://octopart.com/search?q=ES2J-LTP

WHOA hey... I will dig into this. I saw HH post about it and loved it but could never find parts for the voltage I need.
I will revisit it. Dude thanks so much I have developed around the ACPL driver and would like to save the time of learning a new one for now.

-Arlin
 
I see HHs post again...

2 things that scare me first is you could pull the gate on hard and maybe put to much voltage into the gate? I am sure as the voltage drops on the Collector the current flowing into the gate will disappear but still scares me. So Should you not add a resistor in series to make it 3 parts.

Second. I fully charge to 470v the 440v rated part might not work... I will do some digging.

Thanks again. Either way I will add pads for the 3 parts to consider using them.

Edit: Maybe you can put 2 in series? I would then add 4 pads ;)
EDIT MORE: HH was the first to show me as well. At first I thought it was something in the driver controlling this. But nope is a TVS and a fast reverse blocking diode. I just ordered some 350 400 and 440v TVS diodes to experiment with. I have a few things that could use these. Thanks again. And HH was the first to show this to me but I was stumped on how to find the voltage I needed.
 
Yeah I found odd your 300hp drive didnt have this and HH's 10hp had it.
Please let me know which tvs works for you (it will work for me as well) so I can update my BOM.
 
About a gate overvoltage, if it happens the gate itself will turn on the switch and extinguish the overshoot that causes the gate overvoltage. The big chunk of silicon will handle that power better than the gate drive circuitry.

I guess it could happen so fast that your switch doesn't react fast enough to the gate overvoltage, in that case you have shottkys clamping gate voltage to +15/-8v. They will either break or send power to the gate driver VCC. By that time your switch should have been able to handle the overshoot.

From the datasheet, the 440 tvs has a 440v reverse standoff voltage. And at 492V it will let flow 1mA.

This 1mA current into the gate is sunk by the gate pulldown resistors (4.7k pulldown and 1ohm turn OFF resistor)causing a voltage drop not enough to turn on the protection. Some real current flow needs to happen to overcome the low resistance turn off resistor.

In case you wonder, the infineon kit you found uses this tvs
https://www.digikey.com/product-detail/en/littelfuse-inc/P6SMB510A/P6SMB510ALFCT-ND/1766708

Its quite similar to mine, it works at 10v more. Mine has twice the current rating 2.1A vs 0.8A. I still don't understand how this clamps with only 2A, it looks to me like it will let flow like 7A until it overcomes the low resistance turn OFF resistor.
Ah, my TVS withstands 10KW for 10us, so it can sink 14A at 700v. Makes sense now.

file.php
 

Attachments

  • clamping.png
    clamping.png
    26.4 KB · Views: 1,988
Yeah I am using shottkys to clamp the gate to +15 and -5 that I run.
And yeah just after clicking the order button I read and found the part infineon uses...

Ok either way there are on order and I will run some tests when I can.

My gate control has been really good ;) As well I am using a switch rated for 1200v while only charging the car to 470v..
Like I said I think or have a gut feeling that when I push it to hard this over voltage spike over 1200v might have killed a few IGBTs

When filming with discovery channel the car was running perfect but at the track Luke wanted me to turn the power to max at 790 phase amps and hold the car with the brakes and pin the throttle and hold it until the light turns green.

I chose not to do this at the track as I kept it about 25% below max for a launch and turned it up after the car started moving and worked great as I was traction limited anyways. But in front of my shop the next day I pinned it and slowly turned the power to max it lasted about 1/2 a second till a IGBT died and desat tripped and I was pushing it into my shop. Nothing was hot although the die inside the modules might have been hot but my gut feeling is it was a voltage overshoot....

Testing and proving the last .000001% of reliability will cost some money ;) But its worth it!
 
Arlo1 said:
When filming with discovery channel the car was running perfect but at the track Luke wanted me to turn the power to max at 790 phase amps and hold the car with the brakes and pin the throttle and hold it until the light turns green.

I chose not to do this at the track as I kept it about 25% below max for a launch and turned it up after the car started moving and worked great as I was traction limited anyways. But in front of my shop the next day I pinned it and slowly turned the power to max it lasted about 1/2 a second till a IGBT died and desat tripped and I was pushing it into my shop. Nothing was hot although the die inside the modules might have been hot but my gut feeling is it was a voltage overshoot....

Testing and proving the last .000001% of reliability will cost some money ;) But its worth it!
If after 1/2 second of full blocked torque it blows I wouldn't say I'm at 99.99999% reliability :p

In your build can you improve the dc link->switch inductance? The build of this thread would have something like 4nH.

My controller has some discrete logic that differentiates and latches a fault comming from the gate driver (desat) from a fault coming from the current sensor (overcurrent protection). I can tell what happened by looking at the leds.

My active clamping is there to cover the case of a very high current and the gate driver is commanded to turn off just *before* a DESAT event, the IGBT would turn off fast as usual, the Vce overvoltage would be high and no soft turn-off would happen (its only triggered by a desaturation event).

Happy testing!
 
I should add 790 phase amps is much above what I normally run. For the winter it's at 700 and if I push it in the summer usually 750 is max.

I was trying to see how far I could go.

I use a Arduino to flash a led to tell me what stage had the fault. There is 100s of times there is no issue but it trips desat as a over current because of a incorrect setting or something else I did wrong while testing.

Either way with the power turned down to about 200hp it seems to be indestructible but it's only when I push the limits I have problems..

As for changing the DC link switch inductance that requires a new layout. I'm heading there with the big controler. All in time.
 
Back
Top